net/bridge/br_if.c

   1 /*
   2  *      Userspace interface
   3  *      Linux ethernet bridge
   4  *
   5  *      Authors:
   6  *      Lennert Buytenhek               <buytenh@gnu.org>
   7  *
   8  *      $Id: br_if.c,v 1.7 2001/12/24 00:59:55 davem Exp $
   9  *
  10  *      This program is free software; you can redistribute it and/or
  11  *      modify it under the terms of the GNU General Public License
  12  *      as published by the Free Software Foundation; either version
  13  *      2 of the License, or (at your option) any later version.
  14  */
  15
  16 #include <linux/kernel.h>
  17 #include <linux/netdevice.h>
  18 #include <linux/ethtool.h>
  19 #include <linux/if_arp.h>
  20 #include <linux/module.h>
  21 #include <linux/init.h>
  22 #include <linux/rtnetlink.h>
  23 #include <linux/if_ether.h>
  24 #include <net/sock.h>
  25
  26 #include "br_private.h"
  27
  28 /*
  29  * Determine initial path cost based on speed.
  30  * using recommendations from 802.1d standard
  31  *
  32  * Need to simulate user ioctl because not all device's that support
  33  * ethtool, use ethtool_ops.  Also, since driver might sleep need to
  34  * not be holding any locks.
  35  */
  36 static int port_cost(struct net_device *dev)
  37 {
  38         struct ethtool_cmd ecmd = { ETHTOOL_GSET };
  39         struct ifreq ifr;
  40         mm_segment_t old_fs;
  41         int err;
  42
  43         strncpy(ifr.ifr_name, dev->name, IFNAMSIZ);
  44         ifr.ifr_data = (void __user *) &ecmd;
  45
  46         old_fs = get_fs();
  47         set_fs(KERNEL_DS);
  48         err = dev_ethtool(&ifr);
  49         set_fs(old_fs);
  50
  51         if (!err) {
  52                 switch(ecmd.speed) {
  53                 case SPEED_100:
  54                         return 19;
  55                 case SPEED_1000:
  56                         return 4;
  57                 case SPEED_10000:
  58                         return 2;
  59                 case SPEED_10:
  60                         return 100;
  61                 }
  62         }
  63
  64         /* Old silly heuristics based on name */
  65         if (!strncmp(dev->name, "lec", 3))
  66                 return 7;
  67
  68         if (!strncmp(dev->name, "plip", 4))
  69                 return 2500;
  70
  71         return 100;     /* assume old 10Mbps */
  72 }
  73
  74
  75 /*
  76  * Check for port carrier transistions.
  77  * Called from work queue to allow for calling functions that
  78  * might sleep (such as speed check), and to debounce.
  79  */
  80 static void port_carrier_check(void *arg)
  81 {
  82         struct net_bridge_port *p = arg;
  83
  84         rtnl_lock();
  85         if (netif_carrier_ok(p->dev)) {
  86                 u32 cost = port_cost(p->dev);
  87
  88                 spin_lock_bh(&p->br->lock);
  89                 if (p->state == BR_STATE_DISABLED) {
  90                         p->path_cost = cost;
  91                         br_stp_enable_port(p);
  92                 }
  93                 spin_unlock_bh(&p->br->lock);
  94         } else {
  95                 spin_lock_bh(&p->br->lock);
  96                 if (p->state != BR_STATE_DISABLED)
  97                         br_stp_disable_port(p);
  98                 spin_unlock_bh(&p->br->lock);
  99         }
 100         rtnl_unlock();
 101 }
 102
 103 static void destroy_nbp(struct net_bridge_port *p)
 104 {
 105         struct net_device *dev = p->dev;
 106
 107         dev->br_port = NULL;
 108         p->br = NULL;
 109         p->dev = NULL;
 110         dev_put(dev);
 111
 112         br_sysfs_freeif(p);
 113 }
 114
 115 static void destroy_nbp_rcu(struct rcu_head *head)
 116 {
 117         struct net_bridge_port *p =
 118                         container_of(head, struct net_bridge_port, rcu);
 119         destroy_nbp(p);
 120 }
 121
 122 /* Delete port(interface) from bridge is done in two steps.
 123  * via RCU. First step, marks device as down. That deletes
 124  * all the timers and stops new packets from flowing through.
 125  *
 126  * Final cleanup doesn't occur until after all CPU's finished
 127  * processing packets.
 128  *
 129  * Protected from multiple admin operations by RTNL mutex
 130  */
 131 static void del_nbp(struct net_bridge_port *p)
 132 {
 133         struct net_bridge *br = p->br;
 134         struct net_device *dev = p->dev;
 135
 136         /* Race between RTNL notify and RCU callback */
 137         if (p->deleted)
 138                 return;
 139
 140         dev_set_promiscuity(dev, -1);
 141
 142         cancel_delayed_work(&p->carrier_check);
 143         flush_scheduled_work();
 144
 145         spin_lock_bh(&br->lock);
 146         br_stp_disable_port(p);
 147         p->deleted = 1;
 148         spin_unlock_bh(&br->lock);
 149
 150         br_fdb_delete_by_port(br, p);
 151
 152         list_del_rcu(&p->list);
 153
 154         call_rcu(&p->rcu, destroy_nbp_rcu);
 155 }
 156
 157 /* called with RTNL */
 158 static void del_br(struct net_bridge *br)
 159 {
 160         struct net_bridge_port *p, *n;
 161
 162         list_for_each_entry_safe(p, n, &br->port_list, list) {
 163                 br_sysfs_removeif(p);
 164                 del_nbp(p);
 165         }
 166
 167         del_timer_sync(&br->gc_timer);
 168
 169         br_sysfs_delbr(br->dev);
 170         unregister_netdevice(br->dev);
 171 }
 172
 173 static struct net_device *new_bridge_dev(const char *name)
 174 {
 175         struct net_bridge *br;
 176         struct net_device *dev;
 177
 178         dev = alloc_netdev(sizeof(struct net_bridge), name,
 179                            br_dev_setup);
 180
 181         if (!dev)
 182                 return NULL;
 183
 184         br = netdev_priv(dev);
 185         br->dev = dev;
 186
 187         spin_lock_init(&br->lock);
 188         INIT_LIST_HEAD(&br->port_list);
 189         spin_lock_init(&br->hash_lock);
 190
 191         br->bridge_id.prio[0] = 0x80;
 192         br->bridge_id.prio[1] = 0x00;
 193         memset(br->bridge_id.addr, 0, ETH_ALEN);
 194
 195         br->feature_mask = dev->features;
 196         br->stp_enabled = 0;
 197         br->designated_root = br->bridge_id;
 198         br->root_path_cost = 0;
 199         br->root_port = 0;
 200         br->bridge_max_age = br->max_age = 20 * HZ;
 201         br->bridge_hello_time = br->hello_time = 2 * HZ;
 202         br->bridge_forward_delay = br->forward_delay = 15 * HZ;
 203         br->topology_change = 0;
 204         br->topology_change_detected = 0;
 205         br->ageing_time = 300 * HZ;
 206         INIT_LIST_HEAD(&br->age_list);
 207
 208         br_stp_timer_init(br);
 209
 210         return dev;
 211 }
 212
 213 /* find an available port number */
 214 static int find_portno(struct net_bridge *br)
 215 {
 216         int index;
 217         struct net_bridge_port *p;
 218         unsigned long *inuse;
 219
 220         inuse = kmalloc(BITS_TO_LONGS(BR_MAX_PORTS)*sizeof(unsigned long),
 221                         GFP_KERNEL);
 222         if (!inuse)
 223                 return -ENOMEM;
 224
 225         memset(inuse, 0, BITS_TO_LONGS(BR_MAX_PORTS)*sizeof(unsigned long));
 226         set_bit(0, inuse);      /* zero is reserved */
 227         list_for_each_entry(p, &br->port_list, list) {
 228                 set_bit(p->port_no, inuse);
 229         }
 230         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 231         kfree(inuse);
 232
 233         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 234 }
 235
 236 /* called with RTNL but without bridge lock */
 237 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 238                                        struct net_device *dev)
 239 {
 240         int index;
 241         struct net_bridge_port *p;
 242
 243         index = find_portno(br);
 244         if (index < 0)
 245                 return ERR_PTR(index);
 246
 247         p = kmalloc(sizeof(*p), GFP_KERNEL);
 248         if (p == NULL)
 249                 return ERR_PTR(-ENOMEM);
 250
 251         memset(p, 0, sizeof(*p));
 252         p->br = br;
 253         dev_hold(dev);
 254         p->dev = dev;
 255         p->path_cost = port_cost(dev);
 256         p->priority = 0x8000 >> BR_PORT_BITS;
 257         dev->br_port = p;
 258         p->port_no = index;
 259         br_init_port(p);
 260         p->state = BR_STATE_DISABLED;
 261         INIT_WORK(&p->carrier_check, port_carrier_check, p);
 262         kobject_init(&p->kobj);
 263
 264         return p;
 265 }
 266
 267 int br_add_bridge(const char *name)
 268 {
 269         struct net_device *dev;
 270         int ret;
 271
 272         dev = new_bridge_dev(name);
 273         if (!dev)
 274                 return -ENOMEM;
 275
 276         rtnl_lock();
 277         if (strchr(dev->name, '%')) {
 278                 ret = dev_alloc_name(dev, dev->name);
 279                 if (ret < 0)
 280                         goto err1;
 281         }
 282
 283         ret = register_netdevice(dev);
 284         if (ret)
 285                 goto err2;
 286
 287         /* network device kobject is not setup until
 288          * after rtnl_unlock does it's hotplug magic.
 289          * so hold reference to avoid race.
 290          */
 291         dev_hold(dev);
 292         rtnl_unlock();
 293
 294         ret = br_sysfs_addbr(dev);
 295         dev_put(dev);
 296
 297         if (ret)
 298                 unregister_netdev(dev);
 299  out:
 300         return ret;
 301
 302  err2:
 303         free_netdev(dev);
 304  err1:
 305         rtnl_unlock();
 306         goto out;
 307 }
 308
 309 int br_del_bridge(const char *name)
 310 {
 311         struct net_device *dev;
 312         int ret = 0;
 313
 314         rtnl_lock();
 315         dev = __dev_get_by_name(name);
 316         if (dev == NULL)
 317                 ret =  -ENXIO;  /* Could not find device */
 318
 319         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 320                 /* Attempt to delete non bridge device! */
 321                 ret = -EPERM;
 322         }
 323
 324         else if (dev->flags & IFF_UP) {
 325                 /* Not shutdown yet. */
 326                 ret = -EBUSY;
 327         }
 328
 329         else
 330                 del_br(netdev_priv(dev));
 331
 332         rtnl_unlock();
 333         return ret;
 334 }
 335
 336 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 337 int br_min_mtu(const struct net_bridge *br)
 338 {
 339         const struct net_bridge_port *p;
 340         int mtu = 0;
 341
 342         ASSERT_RTNL();
 343
 344         if (list_empty(&br->port_list))
 345                 mtu = ETH_DATA_LEN;
 346         else {
 347                 list_for_each_entry(p, &br->port_list, list) {
 348                         if (!mtu  || p->dev->mtu < mtu)
 349                                 mtu = p->dev->mtu;
 350                 }
 351         }
 352         return mtu;
 353 }
 354
 355 /*
 356  * Recomputes features using slave's features
 357  */
 358 void br_features_recompute(struct net_bridge *br)
 359 {
 360         struct net_bridge_port *p;
 361         unsigned long features, checksum;
 362
 363         features = br->feature_mask &~ NETIF_F_IP_CSUM;
 364         checksum = br->feature_mask & NETIF_F_IP_CSUM;
 365
 366         list_for_each_entry(p, &br->port_list, list) {
 367                 if (!(p->dev->features
 368                       & (NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM)))
 369                         checksum = 0;
 370                 features &= p->dev->features;
 371         }
 372
 373         br->dev->features = features | checksum | NETIF_F_LLTX;
 374 }
 375
 376 /* called with RTNL */
 377 int br_add_if(struct net_bridge *br, struct net_device *dev)
 378 {
 379         struct net_bridge_port *p;
 380         int err = 0;
 381
 382         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER)
 383                 return -EINVAL;
 384
 385         if (dev->hard_start_xmit == br_dev_xmit)
 386                 return -ELOOP;
 387
 388         if (dev->br_port != NULL)
 389                 return -EBUSY;
 390
 391         if (IS_ERR(p = new_nbp(br, dev)))
 392                 return PTR_ERR(p);
 393
 394         if ((err = br_fdb_insert(br, p, dev->dev_addr)))
 395                 destroy_nbp(p);
 396
 397         else if ((err = br_sysfs_addif(p)))
 398                 del_nbp(p);
 399         else {
 400                 dev_set_promiscuity(dev, 1);
 401
 402                 list_add_rcu(&p->list, &br->port_list);
 403
 404                 spin_lock_bh(&br->lock);
 405                 br_stp_recalculate_bridge_id(br);
 406                 br_features_recompute(br);
 407                 if ((br->dev->flags & IFF_UP)
 408                     && (dev->flags & IFF_UP) && netif_carrier_ok(dev))
 409                         br_stp_enable_port(p);
 410                 spin_unlock_bh(&br->lock);
 411
 412                 dev_set_mtu(br->dev, br_min_mtu(br));
 413         }
 414
 415         return err;
 416 }
 417
 418 /* called with RTNL */
 419 int br_del_if(struct net_bridge *br, struct net_device *dev)
 420 {
 421         struct net_bridge_port *p = dev->br_port;
 422
 423         if (!p || p->br != br)
 424                 return -EINVAL;
 425
 426         br_sysfs_removeif(p);
 427         del_nbp(p);
 428
 429         spin_lock_bh(&br->lock);
 430         br_stp_recalculate_bridge_id(br);
 431         br_features_recompute(br);
 432         spin_unlock_bh(&br->lock);
 433
 434         return 0;
 435 }
 436
 437 void __exit br_cleanup_bridges(void)
 438 {
 439         struct net_device *dev, *nxt;
 440
 441         rtnl_lock();
 442         for (dev = dev_base; dev; dev = nxt) {
 443                 nxt = dev->next;
 444                 if (dev->priv_flags & IFF_EBRIDGE)
 445                         del_br(dev->priv);
 446         }
 447         rtnl_unlock();
 448
 449 }